Adaptive discriminative generative model and application to visual tracking

• US 7,369,682 B2
• Filed: 07/11/2005
• Issued: 05/06/2008
• Est. Priority Date: 07/09/2004
• Status: Active Grant

• Current Assignee: Honda Motor Co., Ltd. (Honda Motor Company)
• Original Assignee: Honda Motor Co., Ltd. (Honda Motor Company)
• Inventors: Ross, David, Lim, Jongwoo, Lin, Ruei-Sung, Yang, Ming-Hsuan
• Primary Examiner(s): Mariam; Daniel
• Application Number:
  US11/179,881
• Publication Number:
  US 20060036399A1
• Time in Patent Office: 1,030 Days
• Field of Search: 382/103, 382106-107, 382/159, 382/190, 382/209, 382/22, 382/4, 348/77, 348/135, 348/143, 348/169, 707/6, 707/7
• US Class Current: 382/103
• CPC Class Codes: G06F 18/2132 based on discrimination criteria, e.g. discriminant analysis
• CPC Class Codes: G06F 18/2135 based on approximation criteria, e.g. principal component analysis
• CPC Class Codes: G06T 2207/30241 Trajectory
• CPC Class Codes: G06T 7/246 using feature-based methods, e.g. the tracking of corners or segments
• CPC Class Codes: G06T 7/277 involving stochastic approaches, e.g. using Kalman filters
• CPC Class Codes: G06T 7/70 Determining position or orientation of objects or cameras camera calibration G06T7/80
• CPC Class Codes: G06V 10/25 Determination of region of interest [ROI] or a volume of interest [VOI]
• CPC Class Codes: G06V 10/76 based on eigen-space representations, e.g. from pose or different illumination conditions; Shape manifolds
• CPC Class Codes: G06V 10/7715 Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
• CPC Class Codes: G06V 40/164 using holistic features
• CPC Class Codes: G06V 40/167 using comparisons between temporally consecutive images

Abstract

A system and a method are disclosed for an adaptive discriminative generative model with a probabilistic interpretation. As applied to visual tracking, the discriminative generative model separates the target object from the background more accurately and efficiently than conventional methods. A computationally efficient algorithm constantly updates the discriminative model over time. The discriminative generative model adapts to accommodate dynamic appearance variations of the target and background. Experiments show that the discriminative generative model effectively tracks target objects undergoing large pose and lighting changes.

Claims

1. A computer-based method for classifying an observation of a set of observations, the method comprising the steps of:

(a) receiving the set of observations from a first time period;

(b) classifying members of the set of observations as one of a first and a second type based upon a discriminative-generative model based upon observations prior to said first time period;

(c) modeling a probability density of the set of observations by assigning a first set of probabilities to members of the set of observations classified as said first type and a second set of probabilities to members of the set of observations classified as said second type based upon said discriminative-generative model based upon observations prior to said first time period

(d) revising said discriminative-generative model, to account for said observations from said first time period, based upon said first and said second set of probabilities; and

repeating steps (a)-(d) for a time period after said first time period.

2. A computer-based method for tracking a location of an object within two or more digital images of a set of digital images, the method comprising the steps of:

receiving a first image vector representing a first image within the set of digital images;

determining the location of the object within said first image from said first image vector;

applying a first model to said first image vector to determine a possible motion of the object between said first image vector and a successive image vector representing a second image within the set of digital images;

applying an second model to said first image vector to determine a most likely location of the object within said successive image vector from a set of possible locations of the object within said successive image vector;

applying a third model classifying said successive image vector as one of a first type and a second type;

applying an inference model to said first, second and third models to predict said most likely location of the object; and

updating an Eigenbasis representing an image space of the two or more digital images.

3. The method of claim 2, wherein said first model comprises a dynamic model.

4. The method of claim 2, wherein said second model comprises an observation model.

5. The method of claim 2, wherein said third model comprises a discriminative-generative model.

6. The method of claim 5, wherein said applying a discriminative-generative model further comprises:

(a) receiving a set of observations from a first time period;

(b) classifying members of the set of observations as one of a first and a second type based upon a discriminative-generative model based upon observations prior to said first time period

(c) modeling a probability density of the set of observations by assigning a first set of probabilities to members of the set of observations classified as said first type and a second set of probabilities to members of the set of observations classified as said second type based upon said discriminative-generative model based upon observations prior to said first time period

(d) revising said discriminative-generative model, to account for said observations from said first time period, based upon said first and said second set of probabilities; and

repeating steps (a)-(d) for a time period after said first time period.